U.S. patent application number 10/187758 was filed with the patent office on 2004-01-08 for decision support system using narratives for detecting patterns.
This patent application is currently assigned to International Business Machines Corporation. Invention is credited to Kurtz, Cynthia, Snowden, David.
Application Number | 20040006567 10/187758 |
Document ID | / |
Family ID | 29999397 |
Filed Date | 2004-01-08 |
United States Patent
Application |
20040006567 |
Kind Code |
A1 |
Kurtz, Cynthia ; et
al. |
January 8, 2004 |
Decision support system using narratives for detecting patterns
Abstract
A method for providing a decision support system, has the steps
of: (a) storing in a database a plurality of documents, each
document representing a narrative and each document being tagged
with metadata; (b) searching the database for a narrative that
corresponds to a search criteria of the data in the metadata in the
narrative; and (c) displaying a set of results.
Inventors: |
Kurtz, Cynthia; (Edinburg,
NY) ; Snowden, David; (Marlborough, GB) |
Correspondence
Address: |
IBM CORPORATION
IPLAW IQ0A/40-3
1701 NORTH STREET
ENDICOTT
NY
13760
US
|
Assignee: |
International Business Machines
Corporation
Armonk
NY
|
Family ID: |
29999397 |
Appl. No.: |
10/187758 |
Filed: |
July 2, 2002 |
Current U.S.
Class: |
1/1 ; 707/999.1;
707/E17.095 |
Current CPC
Class: |
G06F 16/38 20190101 |
Class at
Publication: |
707/100 |
International
Class: |
G06F 007/00 |
Claims
1. A method for providing a decision support system, the method
comprising the steps of: (a) storing in a database a plurality of
documents, each document representing a narrative and each document
being tagged with metadata; (b) searching the database for a
narrative that corresponds to a user input search criteria using
said metadata in the narrative; and (c) displaying a set of
results.
2. A method as claimed in claim 1, wherein the metadata comprises
attributes and classifications which convey patterns within the
narrative.
3. A method as claimed in claim 1, wherein the metadata
corresponding to a narrative located in the database is searched in
a context that is relevant to a particular situation.
4. A method as claimed in claim 1 wherein, a tagged narrative is
tagged using a markup language.
5. A method as claimed in claim 1, wherein the user input search
criteria is input to the decision support system using a graphical
user interface.
6. A method as claimed in claim 1, wherein step (c) displays
patterns that occur in the narratives.
7. A method as claimed in claim 2, wherein a metamodel is used to
derive the attributes and classifications.
8. A computer program product comprising computer program code
stored on a computer readable storage medium, which when executed
on a data processing system, instructs the data processing system
to carry out the method as claimed in claim 1.
9. A system for providing a decision support system, the system
comprising: (a) means for storing in a database a plurality of
documents, means for each document representing a narrative and
means for each documents being tagged with metadata; (b) means for
searching the database for a narrative that corresponds to a user
input search criteria using said metadata in the narrative; and (c)
means for displaying a set of results.
10. A system as claimed in claim 9, wherein the metadata comprises
means for attributes and classifications which convey patterns
within the narrative.
11. A system as claimed in claim 9, wherein means for the metadata
corresponding to a narrative located in the database is searched in
a context that is relevant to particular situation.
12. A system as claimed in claim 9, wherein, means for the tagged
narrative to be tagged using a markup language is provided.
13. A system as claimed in claim 9, wherein means for the user
input search criteria is input to the decision support system using
a graphical user interface.
14. A system as claimed in claim 9, wherein means for step (c)
displays patterns that occur in the narratives.
15. A system as claimed in claim 10, wherein means for a metamodel
is used to derive the attributes and classifications.
Description
FIELD OF THE INVENTION
[0001] The present invention relates to the field of decision
support systems and particularly to the use of narratives and
tagging to detect patterns that occur within the narratives.
BACKGROUND TO THE INVENTION
[0002] Since ancient times human beings have told stories for many
purposes, including the transfer of knowledge, values and beliefs.
Stories quickly convey complex messages to diverse audiences. One
of the oldest uses of storytelling is in providing a different
perspective. The very act of listening to a story requires the
willing suspension of one's own perspective to temporarily
entertain another view of the world. By looking at a situation from
a different perspective, people and in particular decision makers
can gain a better understanding of why people behave as they do,
for example the way people in different cultures tend to act under
a particular condition.
[0003] Stories or narratives can be used effectively to aid people
in making decisions, particularly to detect emerging patterns about
complex issues such as are found in corporate strategy and foreign
policy. By detecting these emerging patterns decision makers are
better equipped to deal with complex situations that require
complex decision making techniques. A complex situation can be
described as one in which, the interactions of many identities make
simple predictions impossible, for example in financial market
places.
[0004] Decision support systems exist that aid a person in decision
making but these tend to be for very narrow subject or knowledge
domains for example financial management and mathematical
statistical analysis. These types of decision support systems
therefore can not be utilized across many different knowledge
domains and the knowledge within the decision support system is
tightly scoped and therefore has limited use. Current state of the
art decision support systems do not use narratives as a means for
detecting patterns as until now there has been no mechanism to aid
the retrieval process of such narratives and to provide a mechanism
in which to detect patterns in the narratives in a meaningful
manner to aid decision making.
SUMMARY OF THE INVENTION
[0005] In accordance with the present invention there is now
provided a method for a decision support system, the method
comprising the steps of: (a) storing in a database a plurality of
documents, each document representing a narrative and each document
being tagged with metadata;(b) searching the database for a
narrative that corresponds to a search criteria using the metadata
in the narrative; and (c) displaying a set of results.
[0006] Because a narrative is tagged with metadata that is relevant
to the environment in which the narrative was captured and a
plurality of the tagged narratives are stored in a database, the
database can be searched to find a narrative situated in a
meaningful context to a user. This allows for emerging patterns to
be detected, for example changes in cultural beliefs over time or
mismatches between viewpoints.
[0007] Viewed from another aspect the present invention provides a
system for carrying out the method described above.
[0008] Further, the invention provides a computer program product
for instructing a data processing system to carry out the method
described above.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The invention will now be described, by way of example only,
with reference to a preferred embodiment thereof, as illustrated in
the accompanying drawings, in which:
[0010] FIG. 1, illustrates the decision support system, running on
a data processing system, according to a preferred embodiment of
the present invention;
[0011] FIG. 2, illustrates the steps that the decision support
system carries out, according to a preferred embodiment of the
present invention;
[0012] FIG. 3, illustrates the interface of the decision support
system of FIG. 1, according to a preferred embodiment of the
present invention; and
[0013] FIG. 4, illustrates a narrative tagged with a markup
language as it would be stored in the database of FIG. 1, according
to a preferred embodiment of the present invention;
DESCRIPTION OF THE PREFERRED EMBODIMENT
[0014] A user of a decision support system 105 in FIG. 1, is
provided with a meta model which, allows the user of the decision
support system to approach decision making in a variety of
different contexts. For example the meta model can enable the user
to decide what status they are `at` regarding a particular
situation and to enable the user to select the most appropriate
decision making method to use. The meta model provides different
types of classification that particular problems are characterized
by. Examples of classifications for decision making are knowable
systems, where cause and effect relationships can be discovered
through scientific methods, known systems where cause and effect
relationships are static or predictable, complex systems such as an
organization whose identities and interactions are dynamic and
irreducible and chaotic systems where no connections are found. An
example of the different types of systems is one in which a user
might realize that although they had once been in a stable,
knowable situation with regards to a certain opponent, recent
changes have moved the situation into a more complex space. The
meta model can be further used to consider how different people or
groups might see a current situation given their differing opinions
or world views.
[0015] FIG. 1, illustrates a decision support system 105 running on
a data processing system 100. The decision support system 105
comprises an interface 110 and a database 120, where each document
stored in the database represents a narrative tagged with
metadata.
[0016] The interface 110 of the decision support system 105 can be
understood with reference to FIG. 3. The interface 110 comprises a
graphical user interface used for the input of a search criteria,
the retrieval of one or more tagged narratives requested by the
search criteria and for displaying the results in response to the
search criteria. The search criteria is entered into the interface
110 by selecting criteria from a plurality of selection boxes as
shown in section 375 and a plurality of input boxes as shown in
section 395. The search criteria is built using a combination of
the input from both the selection boxes and the input boxes.
[0017] The results are displayed in a variety of ways. Firstly, by
referring to section 370, the results of the search are displayed
as a plurality of bar charts, each bar chart situated in an
individual cell and each cell has a legend which depicts a category
on a vertical axis 330 and a category on a horizontal axis 315. The
individual bar charts illustrate narrative patterns that occur in
the tagged narratives stored in the database 120. Narrative
patterns are contextual details as to how the narratives were told
and heard in the community, such as the source of the narrative,
motivations of the narrative being told and emotional content of
one or more archetypes in the narrative. In section 375 there are 8
selection boxes which, represent the narrative patterns in the
stored narratives. Each row of selection boxes has a designated
color for example, the first row may have the color blue, which is
shown as a blue box to the right of the first row 395. In this case
the user has selected to see the total number 397 of narratives
patterns that represent the search criteria. The results from the
user selecting this particular search criteria is shown as the
first bar 310 of each individual bar chart in each cell. The first
bar 310 of the bar chart corresponds to the color designated to the
selection box 395, thus the individual colors designated to each
selection box acts as a legend for the individual bar charts. Hence
the interface 110 provides a very quick and easy way to see the
narrative patterns.
[0018] Secondly, the narratives that have been retrieved as a
result of a search are displayed in a list box as shown in section
385. In the example given in FIG. 3, it can be seen that a total of
5 narratives have been retrieved resulting from the search criteria
entered by the user. A user can select any one of the listed
narratives to read and the selected narrative is displayed in
section 390.
[0019] The user can navigate around the interface 110 by using a
standard input device such as a keyboard or a pointing device such
as a mouse for manipulating a screen cursor. In response to a
user's movement from an input device or pointer device, the user
may generate user event signals for example, `drag and drop`, using
the mouse for selecting and manipulating selection boxes, input
buttons and list boxes as is known in the art.
[0020] The interface 110 may be embodied on a variety of different
platforms for example a database package such as, Microsoft Access,
which is a registered trade mark of Microsoft Corporation or Lotus
Approach, which is a registered trade mark of Lotus Corporation, or
using a programming language such as Microsoft C++, Visual Basic or
Java. Microsoft C++ and Visual Basic are both registered trade
marks of Microsoft Corporation and Java is a registered trade mark
of Sun Microsystems.
[0021] The interface 110 of FIG. 1, will now be explained further,
by way of an example with reference to FIG. 3, which illustrates a
list of retrieved narratives as shown in section 385. In this
example the user of the decision support system 105 is trying to
understand why their company is experiencing problems with products
misused by their customers, such that the company is faced with a
public backlash. The user first selects a search criteria which
represents the key points of the narrative that the user wishes to
look at. The user has chosen a search criteria of `Archetypes` 300
and `Themes` 305, by which to navigate the tagged narratives stored
in the database 120. In this example, the narrative patterns that
represent the different search criteria are visible in the
interface 110 as the first bar in each cell 310. It is evident in
this example that the majority of narratives are about the `Apathy`
315, `Rational` 320 and `Principled` 325, archetypes rather than
the remaining archetypes of `Vacillator`, `Super cool` and
`Campaigning` and that narratives about `Misuse` 330 and `Harmful
effects` 335, are more prominent than other themes.
[0022] In the `Apathy-Misuse` cell 345, there are no narratives
from `personal experience` 350 represented as the second bar in the
`Apathy-Misuse` cell, but several in the `historical truth`
category 355 represented as the third bar in the `Apathy-Misuse`
cell, and several narratives that are passed around other
communities as being true and not needing proof of truthfulness.
This illustrates that the staff members at the company from whom
the narratives were collected based their opinions about customer
misuse of their product more on hearsay than on direct experience.
This is an example of pattern detection that can aid decision
making. By using the interface 110 to select a search criteria and
search the narratives stored in the database 120, the user is able
to explicitly compare a current situation to a historical situation
sharing deep similarities to determine if any new insights or
options can be found that were not conceivable before or build a
logical argument from the search results, build a coherent story
from a series of narratives in order to describe a complex
situation or build a counter point argument to look at both
perspectives of an issue.
[0023] The database 120 of FIG. 1, contains a plurality of
documents and each document represents a tagged narrative. The
narratives may be primary narratives which consist of oral history
narratives, anecdotes, narratives derived from collective
storytelling sessions, one to one interviews, anonymous virtual
story telling and anthropological observations. Also secondary
narratives can for example, come from materials such as
documentaries, books, articles, news feeds, research projects,
government inquires, military and police investigations. It is
important to note that a narrative is a story as told by a person
and as such the person is not guided by questions posed to them by
another person wishing to elicitate knowledge. By using narratives
from a variety of different sources the scope of the decision
support system 105 covers a broad knowledge base.
[0024] A variety of database platforms may be used, from a
relational database such as Microsoft Access, which is a registered
trade mark of Microsoft Corporation, a hierarchical database such
as Lotus Notes or any other robust database system, a programming
language that allows the programming of such a system or a markup
language such as Exentsible MarkUp Language (XML). FIG. 4
illustrates an example of a narrative tagged with metadata as it
would be stored in the database 120 of the decision support system
105 of FIG. 1.
[0025] A tag for a narrative consists of metadata, which is
contextual data that is encoded with each narrative in the database
120. The tags are provided by a markup language such as XML. XML is
a universal format for structured documents on the world wide web.
XML is a set of rules for designing text formats that allow data to
be structured using tags (words bracketed by `<` and `>`) and
attributes (of the form name="value") to delimit the metadata. The
metadata defined by the attributes provides a classification for
the metadata (name="archetype"), where the metadata "archetype"
forms part of the classification.
[0026] Referring to FIG. 4, it can be seen what a narrative looks
like when it is tagged with a markup language. The tags begin with
the symbols `< >`, which indicate the beginning of a tag and
end with `</>, which indicate the end of a particular markup
tag. To aid narrative pattern detection, the narrative is encoded
with as much metadata as possible so that a user can view a
narrative in context to the environment in which it was captured.
The markup language first declares a tag called
`<storyEntry>` 400, which identifies it as a narrative. Next,
the markup language declares a tag called <Name> 405 which,
identifies the name of the narrative. The body of the narrative is
contained within the tag <Transcript> 410 and includes the
narrative as told by the storyteller and metadata such as whether
the storyteller laughed, sighed or was particularly emotional. The
metadata tags begin with the <ContextFilterMetadata> tag 415
which identifies that this is where the metadata relating to the
classifications begin.
[0027] The attributes of the markup language that form the
classifications of the metadata are identified by the
<ContextFilterType name="Archetype"> 420,
<ContextFilterType name="Stakeholder"> 425 and
<ContextFilterType name="Theme"> 430. In FIG. 3, the
`Archetype` 300, the `Stakeholder` and the `Theme` 305
classifications are displayed as input buttons on the interface
110.
[0028] Turning back to FIG. 4, a narrative pattern metadata tag 435
is used to tag the questions and answers that detect patterns in
the narratives. A question is tagged to the narrative as
<Question text="Emotion level of teller"> 440 and the answer
is tagged to the narrative as <Answer text="low"/> 445.
Referring back to FIG. 3, the narrative that the user has selected
to view in section 385, is the same narrative that can be seen with
its metadata tags in FIG. 4.
[0029] The narrative pattern metadata that is tagged to a narrative
is derived by using a large set of nested questions that a person
could answer about a particular narrative. For example, this could
be questions about the characters in the narrative, the plot, the
setting, emotional level of the story teller and why the narrative
was told. Subsets of metadata are chosen to describe a particular
narrative depending on the amount of information available about
its context. For example, a narrative submitted to a web discussion
group will have a small subset of metadata associated with it
whereas, a narrative captured during a group discussion will have a
larger subset of metadata associated with it. A subset of questions
are selected and the narratives are tagged with the answers to the
questions. The narratives are tagged again with the metadata
attributes and classifications and the tagged narratives are stored
in the database 120. The attributes and classifications metadata
tags are derived from the interaction of diverse groups of people,
for example, people from different parts of society by attending a
workshop. A workshop is a collection of people with a facilitator,
who directs the group through different scenario's or activities
using a metamodel. The group of people attending the workshop
follow the meta model directed by the facilitator to consider,
compare, merge and construct contemporary and historical narratives
with respect to a subject that they have been asked to explore for
example, using the example in FIG. 3, why a customer's product is
being misused. The group of people are guided through a process of
discarding particular biased and stereotypical first class
classifications and arrive at recognizable classifications for
example a classification of an archetype character might be a
`staunch defender` or a `big bully`.
[0030] The attributes of the metadata form an indexing mechanism,
which is based on two factors. The first factor is socially
relevant classifications consisting of, for example, archetypal
characters or situations. The classifications provide a means by
which users of the decision support system 105 can view narratives
as connected to other related categories rather than in isolation.
The second factor is information that captures as much detail as
possible regarding the context in which each particular narrative
was told and heard, for example its source, emotion and motivation
and provides a means by which users of the decision support system
can view narratives which are not torn out of context but situated
in context, so that the user can judge a particular situation side
by side with many other narratives concerning the same
situation.
[0031] Assignment of the markup language questions, answers,
classification and attributes is carried out by either a) manually
tagging a narrative with metadata or b) tagging narratives with
metadata which are suggested or supplied by neural networks (a well
known artificial intelligence technique). Neural networks are used
to suggest answers to a selection of the markup language questions.
The neural network is supplied with one or more narratives that are
already tagged and this is used as training material for the neural
network. The neural network is then able to suggest answers to some
of the questions such as the emotionality of a narrative teller and
audience, purpose of the narrative telling, relationship of the
narrative. The narrative is then tagged with the metadata suggested
by the neural network.
[0032] The tagging of the narratives is normally carried out before
the tagged narratives are stored in the database 120 of the
decision support system 105 of FIG. 1, however, the narratives
could equally be stored in the database 120 of FIG. 1, without
being tagged and then tagged at a later date for use in the
decision support system 105.
[0033] With reference to the flowchart of FIG. 2, it will be
explained how the interface 110 and the database 120 of FIG. 1
interact with each other. The user builds a search criteria from
the plurality of selection boxes 375 and input buttons 395 of FIG.
3 and the search criteria is entered into the decision support
system at step 205. A search string is then formed with the search
criteria that was input into the decision support system 105 from
the interface 110. The database 120 takes the search string and
searches the stored narratives at step 210, using the indexing
mechanism to find and retrieve one or more narratives that match
the search criteria. Once one or more narratives are found that fit
the search criteria, the results are retrieved and the results are
displayed at step 215.
[0034] By using the example in FIG. 3, about customer misuse of
products, the steps that the decision support system 105 carries
out can be further understood. The user selects a search criteria
(at step 205), in this example, the user has chosen to look only at
narratives that feature the `Apathy` archetype and the `Misuse`
theme in which the storytellers (known in this case to be staff
members) have represented the narratives as `historical truth`. The
interface 110 takes the search criteria as the input to the
decision support system 105 and forms a search string therefrom.
The database 120 uses the search string and searches the database
(at step 210) to find and retrieve one or more narratives that
match the search criteria. Once one or more narratives are found,
the results are retrieved and the results are displayed (at step
215) in display areas 370 and 385 of FIG. 3. The narratives are
shown in order of how strongly they match the search criteria. The
user can select a particular narrative from the list of narratives
meeting the selection criteria, and can read the story in order to
gain a deeper understanding of the topic at hand (how staff members
talk about customers who are apathetic and misuse the product). An
audio file and video file of a narrative can be called up and
reviewed, as can its entire set of metadata.
[0035] Further meta models are used in conjunction with the
decision support system 105 to help the user understand the
decisions that the user is required to make. Diagnostic meta models
are used to remove any bias and limited perspective from a decision
making processes for example trying to understand the perspective
of a teenager who constantly reoffends.
[0036] Users are able to internalize the diagnostic meta models by
using them in various exercises designed to help them become
familiar with the use of the models for critical thought and
decision making. These exercises might be individual and or
collective or virtual and or physical. For example users might be
placed in a micro world training session where the users are given
a situation in which they are asked to evaluate or simulate
historical problems and derive options from the situations that the
users feel that they could use if faced with having to make a
decision in such a situation.
[0037] Users are able to understand the diagnostic meta models and
the conceptual underpinnings of the diagnostic meta models by
reading and viewing multimedia explanations about critical concepts
as well as attending physical and virtual training sessions.
[0038] User are able to contextualise the diagnostic meta models to
their unique needs and an environment that they are exploring by
making sense of them within the context of the decision that they
are faced with. For example a user may be asked to explain
historical cases from a piece of literature or from their own
experience in the terms of the diagnostic meta models and arrive at
alternative endings that could be chosen.
[0039] Users can apply the diagnostic meta models in their practice
of decision making by using them as `action filters` for choosing
different decision making activities for example choosing among
such methods as scenario planning, forecasting, simulation, role
playing, logical argument building and statistical analysis. The
user is further able to link particular details of a situation of
concern to relevant abstract concepts and conditions in the
diagnostic meta models thus describing the situation in ways that
increase the user's understanding of the situation for example
finding shift in power structures as played out in recent events,
looking for boundary conditions between states and movements over
boundaries for example, thinking about what would happen if market
reforms led to an unemployment rate increasing beyond the buffer
capacity of an unofficial economy. The user is able to deliberately
consider a situation from more than one perspective or alternative
explanation in order to disrupt entrained thinking about a cause of
a situation for example looking at a recent treaty offer as if it
were intended to mislead to see if looking at it in this manner
might reveal something that has been overlooked before.
* * * * *